Swarm and Evolutionary Computation ()

Contents lists available at SciVerse ScienceDirect

Swarm and Evolutionary Computation

journal homepage: www.elsevier.com/locate/swevo

Regular Paper

Memetic algorithm for minimum energy broadcast problem

in wireless ad hoc networks
D. Arivudainambi n, D. Rekha
Department of Mathematics, Anna University, Chennai, India

art ic l e i nf o

a b s t r a c t

Article history:
Received 19 June 2012
Received in revised form
3 October 2012
Accepted 1 April 2013

The problem of nding a broadcast tree with minimum power consumption has received much attention
due to two reasons. They are (a) the limited bandwidth of wireless networks and (b) the nodes are
battery operated. This work proposes heuristic memetic algorithm to nd minimum energy broadcast
tree in wireless ad hoc networks. The simulation results on numerous problem instances conrm that the
proposed algorithm signicantly outperforms several heuristic algorithms in terms of solution quality.
The experimental evaluation of the memetic algorithm shows that it generally improves over other
recently proposed algorithms. By solving well-known benchmark problem instances with 20 and 50 nodes,
it also demonstrates the effectiveness of memetic algorithm in terms of computation time.
& 2013 Elsevier B.V. All rights reserved.

Keywords:
Wireless ad hoc networks
Minimum energy broadcast
Memetic algorithm

1. Introduction
Ad hoc network is a self-organizing multi-hop wireless network, which relies neither on xed infrastructure nor on predetermined connectivity. Wireless ad hoc networks are used
commonly in the military and emergency situations because of
their quick, easy setup and robust properties.
In a wireless ad hoc network, each wireless node has an
omnidirectional antenna and other nodes within the transmission
range of the node will receive the message. A wireless node can
receive the signal from another node if it is within the transmission range of the sender else communication is through multi-hop
wireless links by using intermediate nodes to relay the message. If
transmission range of a wireless node is high, then it can
communicate directly with more neighbors, but it takes more
energy for power consumption. The total transmission power can
be reduced by using intermediate nodes. One major constraint of
ad hoc networks is the limited energy, since the nodes are battery
operated. Frequent recharging or replacement of batteries is not
possible in such networks. This makes energy efciency an
important metric in ad hoc networks.
In this paper, we present heuristic memetic algorithm for the
minimum energy broadcasting (MEB) problem in wireless ad hoc
networks. Broadcasting in wireless networks is different from broadcasting in wired networks because every node in wireless networks
is equipped with omnidirectional antennas and several nodes are
covered in a single transmission. Each node within the transmission

Corresponding author. Tel.: +91 44 22358520.

E-mail address: arivu@annauniv.edu (D. Arivudainambi).

range of the sender in wireless network can receive the broadcast

without any additional cost to the sender. This feature is known as
wireless multicast advantage (WMA). One way of broadcasting in
wireless networks is by adjusting the transmission power of the
source node such that its transmission can reach the farthest node
from it, which will cover all other remaining nodes because of the
WMA. A broadcast scheme is energy efcient, so the main objective
is to construct a minimum-energy broadcast tree rooted at the
source node with minimum energy consumption.
Wieselthier et al. [6,7] proposed algorithms to nd energy
efcient broadcast and multicast trees in wireless networks.
They proposed broadcast incremental power (BIP) algorithm and
adapted it for multicast operation by introducing the multicast
incremental power algorithm, for some of the fundamental issues
associated with energy efcient broadcasting and multicasting in
wireless networks. The problem they addressed involves the
designation of which nodes are to transmit and at what power
levels. These algorithms exploit the broadcast nature of the
wireless communication environment and address the need for
energy efcient operation.
Das et al. [2] have proposed the r-shrink procedure, which is a
simple local search heuristic to improve the broadcast trees in
wireless networks. After generating initial broadcast trees,
r-shrink procedure is carried out to minimize the power consumption. Given an initial broadcast tree, the transmission radii of
each transmitting non-leaf node in the tree is shrunk sequentially
and the algorithm nds a better position to accommodate the
nodes, which have been disconnected from the tree as a result of
the shrinkage operation. This process is repeated until no further
improvement is possible. If the value of r is 1, then only one node
is removed. Similarly if r is 2, then two nodes are removed.

2210-6502/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.swevo.2013.04.001

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

A hybrid approach combining genetic algorithm operations

cycle crossover and swap mutation, with local search heuristic
for this problem is proposed by Singh and Bhukya [1]. In the local
search operation they have implemented two variations of
r-shrink procedure. The genetic algorithm with 1-shrink procedure
is given as GA-1. The combination of genetic algorithm with
1-shrink and 2-shrink procedure is given as GA-2. Compared to
GA-1, GA-2 identies the solution the maximum number of times
out of 30 runs.
An integer programming model and heuristic algorithm was
proposed by Yuan et al. [4]. This heuristic algorithm does not perform
exhaustive search, they rather execute a predened number of search
moves by simply adopting a maximum time limit.
Wu et al. [14] proposed permutation encoded genetic algorithm to nd a minimum energy broadcast tree. They formulated
the problem mathematically to a constrained optimization problem using graph representation and then genetic algorithm based
approach is developed to obtain the broadcast tree.
Kang and Poovendran [5] proposed an iterated local optimization (ILO) technique at the cost of additional computational
complexity to solve the minimum energy broadcast problem over
wireless ad hoc networks. Montemanni et al. [10] have presented
a new heuristic algorithm based on the simulated annealing
paradigm for solving the minimum power broadcast problem in
wireless networks.
Al-Shihabi et al. [11] studied a new hybrid heuristic algorithm
to solve MEB problem by combining nested partitioning with local
search and linear programming. The major part of this work is
the nested partitioning, where a generic partitioning scheme was
implemented.
A new evolutionary local search for the minimum energy
broadcast problem by Wolf et al. [13] used a modied r-shrink
procedure as local search and random increase in transmission
power of some nodes as mutation operator. The algorithm has
shown to nd optimal or near-optimal solutions in reduced
computation time for the considered test instances. A detailed
survey on the MEB problem for wireless ad hoc networks with
omni-directional and directional antennas can be found in [12].
MEB problem for an antenna model that is realistic for sensor
networks is proposed in [16]. The classical MEB problem is
formulated for antennas where the transmission power can be
adjusted to any desired real value between 0 and the maximum
transmission power. In this paper, an antenna model with transmission power chosen from a nite set of discrete values is taken
for consideration.
A distributed ant colony optimization algorithm for solving the
MEB problem in sensor networks with realistic antennas is
proposed in [17] based on the classical BIP heuristic. A new
localized criterion for extending partial solutions during solution
construction is introduced in order to make use of the mechanism
of BIP.
Enan et al. [18] proposed an evolutionary based clustering
protocol for the routing problem in wireless sensor networks,
where all the transmissions are based on single-hop communication. They formulated a new objective function that can have a
signicant impact on the overall performance of the wireless
sensor networks. It is modied to meet the contradictory goals,
maximal stability period until rst node dies, maximal network
longevity until last node dies, at the same time minimizing energy
consumption throughout the network lifetime.
Sengupta et al. in [19] proposed an energy efcient sensor
manager for tracking of dynamic objects through differentiated
coverage. They developed a technique for tracking of moving
objects in an area that changes their position with time through
differentiated coverage. The sensor management process consists
of tracking and energy efcient optimization. They mainly perform

the operation of sensor manager through optimization by evolutionary algorithm. The process is to obtain an energy efcient
tracking of the moving group so as to extend the net lifetime of the
sensor network.
In [20] they have considered the deployment of sensor nodes
in a given area with the following objectives, (i) minimizing
the number of sensor nodes to reduce cost of deployment,
(ii) minimizing the net energy consumed by all the nodes,
(iii) maximizing the area covered by the nodes, (iv) maximizing
the lifetime of the network. They formulated the sensor node
deployment task as a constrained multi-objective optimization
problem. They developed multi-objective evolutionary algorithm
that uses a new fuzzy dominance based decomposition approach.
For the minimum energy broadcast problem, the approaches
analyzed above suffered from the trade-off between solution
quality and running time. Computation time is an important factor
to validate an algorithm. Therefore, an algorithm is still needed for
broadcast scheduling problem that improves the solution quality
in reduced computation time even for a large network. The
objective of this work is to nd optimal or near optimal broadcast
tree in an acceptable execution time.
Genetic algorithms (GAs) work successfully to solve many
search and optimization problems. However, they may drop into
local optimal solutions or they may nd the optimal solution by
low convergence speed and GA blindly wanders over the search
space. To overcome these problems, we used Memetic algorithm
(MA) to enhance the GA. MA proposed in this study reduces the
processing time nearly 50% when compared very recent algorithm
in [1]. A series of simulations is conducted to evaluate the
performance of the proposed MA in terms of solution quality
and running time, and to verify its superiority over other recent
heuristic algorithms.
MA establishes local search techniques at specic parts of the GA
optimization process, with an aim to increase its performance. It is a
blooming dialect, mainly due to their success in solving many hard
optimization problems. A particular feature of MA is greatly responsible for this unlike traditional evolutionary computation (EC)
methods. MA is intrinsically concerned with exploiting all available knowledge about the problem under study this is something
neglected in evolutionary algorithms (EA) for a long time.
In addition to Darwinism, MA adopts the Lamarckian theory
that offspring can inherit the knowledge or characteristics that
their parents acquire during their lifetime. The MA implements
this idea by integrating a local enhancement, such as local search
and repair operator, into the canonical EA, and making the
enhancement inheritable, this integration signicantly improves
the exploitation ability of EA. In genetic algorithm, the mutation
creates new genes for the population and the crossover operator
orients seeking the best solution from the genes in the population.
In MA, this orientation is achieved by local search. Local search
reduces the search space and reaches to the high quality solution
faster. MA actively aims on improving solution and explicitly
concerned with exploiting all available knowledge about the
problem.
The rest of this paper is organized as follows: Section 2 gives a
formal denition of the problem, along with the constraints. In
Section 3, we describe the algorithm and its operators. The details
of simulation results, computation time by MA with other competitive algorithms are discussed in Section 4. Finally, conclusions are
drawn in Section 5.

2. Minimum energy broadcast problem

Broadcasting is a method to allow all nodes to share the
data efciently with all the other nodes in the wireless networks.

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

Due to the limited energy resources, energy efciency is a fundamental aspect in constructing the broadcast trees. Considering a
set of nodes in which one node is assigned as source node, the
MEB is to minimize the total power consumption when the nodes
in the networks are connected together and they communicate
with the other remaining nodes. In this, the nodes have the ability
to adjust their transmission power. Accordingly, each node is
assigned a transmission range and every other node inside this
range receives the message. The aim is to assign ranges in such a
way that the total energy consumed is minimized.
The MEB problem can be stated using graph theoretical terms.
Let G (V,E) be a directed complete graph, where V denotes the set
of nodes i.e., the terminals of the wireless network and E denotes
the set of edges. The power required to establish a link from node i
to j is given as pij where (i,j)E. The broadcasting property of
wireless networks where nodes are equipped with omnidirectional antennae is given as, if node x is transmitting directly to
node y, each other node i such that pxipxy will also be reached by
the signal transmitted by node x. The MEB problem is to nd a
broadcast routing tree with minimum transmission power such
that a source node sV that has to broadcast a message to
remaining nodes of V either directly or through intermediate
nodes. The start node of each directed edge of a tree is responsible
for relaying the broadcast message to the terminal node of the
same edge. The transmission energy required by node x is
determined by the power required to transmit to the farthest
child of x in that tree. The transmission energy required by leaf
nodes is zero, since leaf nodes are not relaying messages to any
other node. The total transmission energy required for the broadcast tree is computed by adding the energy required by each
parent node in that tree. The MEB problem is to nd a tree rooted
at s with minimum total energy required. It is calculated as,
TP soli max dp; q
pV p;qEsoli

Minimum energy broadcast trees are generated after each

iteration by the algorithm, which is called as solution. For each
solution the total transmission power generated by soli is TP soli ,
where i is the solution number index, is a constant referred to
the path loss exponent that is between 2 and 4, d(p,q) is the
distance between the nodes p and q. For a transmission from node
p to q, the received signal power at q varies as d(p,q) where
h
i12 
dp; q xp xq 2 yp yq 2
2
is the Euclidean distance between nodes p and q, (xp,yp) are the
coordinates of node p and (xq,yq) are the coordinates of node q. The
optimal or near optimal solution is the one, which contains
minimum total transmission power minTP sol :

3. Proposed memetic algorithm

A population based search algorithm called Genetic algorithm
is commonly used to solve combinatorial optimization problems
where the goal is to nd the best solution in a (possibly
unbounded) solution space. It uses the principle of biological
evolution to generate better solutions from previous generations
of solutions. MAs are extensions of evolutionary algorithms that
apply local search processes in the agents and trying to improve its
tness. MA was introduced in the late 80s to denote a family
of metaheuristics. MA exploits all the available problem knowledge by incorporating pre-existing heuristics, preprocessing data
reduction rules, local search techniques, specialized recombination
operators, truncated exact methods, etc. In addition, an important
factor is the use of adequate representations of the problem being

tackled. Unlike other approaches, MA is concerned in exploiting all

available knowledge about the problem under study, which is not
taken as an optional method, but as a fundamental characteristic.
MA is able to improve convergence time, hence making it more
favorable over GA. This results in highly efcient optimization
tools [3,8,9].
In this section, we present the framework of the proposed MA.
The key contribution in our algorithm is to nd the optimal or near
optimal solution tree with reduced computational time. Compared
to other approaches, MAs are superior, because of wide applicability. Despite the good results obtained by some MAs, the
process of designing efcient MA often depends on the problem
specic details. The construction of MA to minimum energy
broadcast tree is given as follows.
Memetic algorithm
Calculate the power matrix;
Generate initial population;
While (until termination condition)
{
Knowledge added Mutation;
Perform 1-shrink;
Perform 2-shrink;
Fitness evaluation;
Survival (initial population, new best population);
}

The MA starts by generating the power matrix of the given

network using Eq. (2) based on the randomly generated x and y
coordinates of the n nodes.
3.1. Initial population
A decoder is a method used to produce the initial population
tree, which is similar to the one given by Wu et al. [14].
A permutation list with n nodes is created having the source node
as the rst one in the list. The decoder begins by making
the source node as the root of tree, which is the rst node of
the permutation list and by setting its transmission power equal to
the transmission power required by it to transmit to the second
node in the permutation list. The second node along with all those
nodes, that are implicitly covered by this transmission power, are
made the children of the root node in the tree. Those nodes that
were inserted in the partially constructed tree are deleted from the
permutation list having the same order of remaining nodes. The
next node x in the permutation list is added to one of the current
children y of the root node s that has minimum transmission
power to x. Along with x, the remaining nodes in the permutation
list that are within this transmission range is also inserted to y.
This process is continued until all nodes in the permutation list are
added to the newly constructed tree. Each constructed tree is
called as a chromosome or solution and the population is a set of
solutions, each solution being represented by a different tree
constructed knowing the power assigned to its nodes.
3.2. Knowledge added mutation
The major drawback of decoder operator is the insertion of the
second node to the root node. A node z in the second place of
the permutation list is inserted as the children of root node s.
The transmission range from s to z might be large compared to the
transmission power of one of the remaining node to z, which
increases the total transmission power of the chromosome.
To overcome this, knowledge added mutation operator is performed

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

on the selected parents. The mutation operator nds best position

for z in the current tree that reduces total transmission power of the
newly generated tree compared to the previous one, if not the
previous tree is retained.
The generated power matrix for the 10-node network calculated using Eq. (2) is shown in Fig. 3. The x and y coordinates are
generated randomly for this network. The decoder generates an
initial broadcast tree for a 10-node network as shown in Fig. 1 and
its total transmission power is 15.59. Node 3 is inserted to the
source node 10 as it is the second node in the permutation list, but
this is not an appropriate place for node 3 and it also increases the
total transmission power of this tree. By applying the knowledge
added mutation operator on this tree it identies that node 3 can
be best accommodated under node 1 as shown in Fig. 2, which
reduces the total transmission power to 12.49.

non-leaf node. When performing knowledge added mutation

operator for node 3 before r-shrink procedure the algorithm converges to the optimal or near optimal solution as early as possible.
The tness of each population is evaluated using Eq. (1) and
best chromosomes are selected that have minimum transmission
power. The survivor selection follows the Darwinian principle of
survival of the ttest. Only the ttest individuals are selected as
parents for the next-generation. Idea of elitism is to retain some of
the best individuals in each generation. In this study, a small
percentage of best tness individuals are retained to the next
generation. It increases the performance of algorithm, by preventing the loss of best found solution. From each generation 10% of
best solution retained to the next iteration.
The loop or generations are carried out until it reaches the
maximum iteration number or no improvement in the solutions is
found in last 10 iterations.

3.3. r-shrink procedure

4. Simulation results
We have taken two variations of r-shrink procedure given by
Das et al. [2]. The population generated from knowledge added
mutation is carried out with 1-shrink operation. The nodes are
checked for improvement from each non-leaf node and once an
improvement is found then the tree is updated with the new ones
and continues to search until all unexplored nodes are checked for
improvement. If 1-shrink fails to improve the solution, 2-shrink
begins. The procedure stops only when complete cycle of both
1-shrink and 2-shrink fails to improve the solution.
While applying the r-shrink procedure in Fig. 1 the improvement for node 3 will be done at the third level starting from the

Fig. 1. Initial broadcast tree for 10-node network.

A series of simulations are carried out to evaluate the performance of MA. The algorithm is implemented in Matlab on
Windows XP/Intel Core 2 Duo T6600 2.2 GHz machine. MA is
tested with randomly generated 10, 25, 50, 75 and 100 nodes in a
5  5 grid as given by Das et al. [2]. For each instance, the
algorithm is carried out for 50 times and the results are given in
Table 1. In addition, a set of 60 benchmark instances [15] are
tested with MA and the results are compared with other recently
proposed algorithms. Each instance consists of n nodes and one of
the n nodes is designated as the source node. For all the experiments the simulation is carried out with parameters 2, population size is 50 i.e., 50 chromosomes and maximum number of
generations is 50.
The simulation results of MA for varying number of nodes are
given in Table 1. The nodes are randomly generated i.e., for the
rst instance the 10-node network is randomly generated once
and it is executed 50 times. For all the 50 times of execution, the
solution found i.e., the generated transmission power is same so
there is no deviation. The deviation of results is given as excess
value in percentage. Excess value is calculated using Eq. (3). From
the excess value it can be identied how much the remaining
solutions are deviated from the optimal solution.


1 TI
Best i
1  100%
3

TI i 1 Optimum
The variable TI represents the number of times executed for
each node, Besti is the solution found at ith execution, Optimum is
the optimum solution. The excess percentage calculated for one
instance of 25-node network is given in Table 2.
For 10, 25 and 75 node network the optimum solution is found
for all the 50 out of 50 instances i.e., we got the minimum
transmission power for all the 50 times. For 50-node network,
the optimal solution identied 48 times out of 50 instances are the
same and the average excess percentage for the two instances
calculated is 0.16, i.e., for two instance the solution is deviated
from others. For 100-node network, the optimal solution is found
Table 1
Simulation results of MA.
Instance

Number of nodes

Excess (%)

Solution found

Time (s)

1.
2.
3.
4.
5.

10
25
50
75
100

No excess
No excess
0.16
No excess
0.31

50 out of 50 instances
50 out of 50 instances
48 out of 50 instances
50 out of 50 instances
47 out of 50 instances

0.08
0.51
6.02
7.21
14.39

Fig. 2. Broadcast tree after knowledge added mutation process.

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

partitioning (NP) [11] and evolutionary local search (ELS) [13] in

Table 3 for a set of 30 instances with 20 nodes and for another set
of 30 instances with 50 nodes. Number of times the optimal
solution found out of 30 runs, the average excess in percentage
over optimal/best found solution is calculated and average
execution time in seconds is reported in Table 3. The output of
MA validates its efciency in terms of excess percentage and
computation time.

47 times out of 50 instances and the average excess percentage for

the remaining three instances calculated is 0.31. From the results
of Table 1, it is clear that for the remaining instances also, the
transmission power found is not much deviated from the optimum
solution.
Two benchmark instances available in [15] are taken for study
and the results are shown in Tables 35. We have compared MA
with GA-1 and GA-2 [1], iterated local search (ILO) [5], nested

Table 2
Excess percentage calculation.
Instance

Best solution found out of 50

instances (A)

Solution found in the rst

instance (B)

Excess
percentage

25node
network

399074.64

423894.57

B/A 1.06219370

C1.00 0.06219370

D  100 6.21937

6.22%

Table 3
Comparison of MA with other recently proposed algorithm.
Instance

Parameter

MA

GA-1

GA-2

ILO

NP

ELS

20-node network

1. Excess (%)
2. Solution found
3. Time (s)

0
30/30
0.51

0
30/30
0.56

0
30/30
0.86

1.44
18.2/30

0.06
28.7/30
0.33

0.42
28.1/30
0.43

50-node network

1. Excess (%)
2. Solution found
3. Time (s)

0.19
24/30
6.02

0.81
19.27/30
7.49

0.25
22.9/30
13.71

3.89
5.23/30

4.03
5.1/30
9.5

2.50
7.53/30
5.5

Table 4
MA results for MEB problem instances with 20 nodes.
Instance Optimum

p20.00
p20.01
p20.02
p20.03
p20.04
p20.05
p20.06
p20.07
p20.08
p20.09
p20.10
p20.11
p20.12
p20.13
p20.14
p20.15
p20.16
p20.17
p20.18
p20.19
p20.20
p20.21
p20.22
p20.23
p20.24
p20.25
p20.26
p20.27
p20.28
p20.29
Average

407 250.81
446 905.52
335 102.42
488 344.90
516 117.75
300 869.14
250 553.15
347 454.08
390 795.34
447 659.11
316 734.39
289 200.92
314 511.98
346 234.51
301 426.68
457 467.93
484 437.68
380 175.41
320 300.23
461 267.52
403 582.74
271 958.28
328 659.78
326 654.08
395 859.67
453 517.28
461 547.18
389 057.00
279 251.95
299 586.76

ELS

NP

ILO

Excess (%) Found

Time (s) Excess (%) Found

0.05

1.57

0.004

10.62
0.25

0.23

26/30
30/30
30/30
30/30
30/30
30/30
17/30
30/30
30/30
27/30
30/30
30/30
1/30
28/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
24/30
30/30
30/30

0.27
0.42
0.53
0.67
0.45
0.41
0.24
0.49
0.68
0.60
0.48
0.30
0.50
0.36
0.63
0.25
0.67
0.46
0.43
0.63
0.40
0.23
0.22
0.41
0.71
0.22
0.39
0.40
0.32
0.26

0.16

0.38

1.03

0.21
0.02

30/30
30/30
30/30
27/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
17/30
30/30
21/30
30/30
30/30
18/30
29/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30

0.42

28.1/30 0.43

0.06

28.7/30 0.327

GA-1

Time (s) Excess (%) Found

0.30
0.36
0.41
0.46
0.43
0.35
0.18
0.31
0.46
0.41
0.40
0.24
0.20
0.26
0.43
0.24
0.41
0.39
0.31
0.34
0.35
0.19
0.19
0.28
0.42
0.21
0.32
0.38
0.28
0.31

GA-2

MA

Found Time (s) Found Time (s) Found Time (s)

7.09
3.52

0.40
3.19

0.37
0.64

1.93

4.71
1.74
12.63
3.72

2.08

0.07
1.13

30/30
5/30
1/30
30/30
0/30
0/30
30/30
30/30
26/30
0/30
30/30
0/30
30/30
30/30
0/30
0/30
0/30
5/30
30/30
30/30
30/30
30/30
30/30
30/30
0/30
30/30
29/30
0/30
30/30
30/30

30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30

0.45
0.60
0.56
0.76
0.54
0.43
0.48
0.60
0.71
0.61
0.61
0.40
0.50
0.41
0.61
0.46
0.74
0.64
0.49
0.67
0.61
0.54
0.46
0.56
0.65
0.45
0.55
0.54
0.56
0.65

30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30

0.60
0.79
0.80
1.19
0.81
0.59
0.70
0.80
1.35
1.01
1.04
0.61
0.62
0.90
1.18
0.69
1.20
1.01
0.66
1.20
0.81
0.67
0.57
0.78
1.07
0.62
0.96
0.83
0.89
0.89

30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30
30/30

0.41
0.56
0.51
0.67
0.40
0.41
0.45
0.48
0.67
0.58
0.61
0.43
0.44
0.63
0.65
0.47
0.65
0.61
0.42
0.67
0.40
0.43
0.40
0.46
0.65
0.41
0.49
0.45
0.48
0.43

1.44

18.2/30 30/30

0.56

30/30

0.86

30/30

0.51

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

Instance

Optimum

ELS
Excess (%)

p50.00
p50.01
p50.02
p50.03
p50.04
p50.05
p50.06
p50.07
p50.08
p50.09
p50.10
p50.11
p50.12
p50.13
p50.14
p50.15
p50.16
p50.17
p50.18
p50.19
p50.20
p50.21
p50.22
p50.23
p50.24
p50.25
p50.26
p50.27
p50.28
p50.29
Average

399 074.64
373 565.15
393 641.09
316 801.09
325 774.22
382 235.90
384 438.46
401 836.85
334 418.45
346 732.05
416 783.45
369 869.41
392 326.01
400 563.83
388 714.91
371 694.65
414 587.42
355 937.07
376 617.33
335 059.72
414 768.96
361 354.27
329 043.51
383 321.04
404 855.92
363 200.32
406 631.51
451 059.62
415 832.44
380 492.77

ELS (2-shrink)
Found

Time (s)

Excess (%)

Found

ILO
Time (s)

Excess (%)

5.60
1.33
3.21
4.53
3.51
2.25
0.99
5.55

2.89
1.78
3.68
1.51
1.43
0.37
0.36
2.38
5.49
1.34
8.06
0.66
1.72
0.15
1.89
0.81

6.62
4.35
1.52
0.34

0/30
0/30
3/30
0/30
6/30
4/30
8/30
2/30
30/30
0/30
4/30
1/30
1/30
2/30
1/30
0/30
7/30
0/30
0/30
11/30
0/30
23/30
29/30
11/30
5/30
30/30
0/30
0/30
5/30
17/30

4.8
5.7
5.8
5.3
4.7
5.1
6.5
6.5
3.4
8.8
5.1
3.1
8.1
8.4
2.4
5.4
9.0
3.6
8.2
4.8
5.2
5.7
2.7
8.7
4.9
1.3
4.9
4.5
9.0
3.4

0.41
0.16
0.28
1.71
0.30
0.83

0.54

3.29
1.16
2.87
0.57
0.04
0.34
0.20
0.30
1.88
0.24

0.15

0.07

2.17
0.18
0.47
0.08

15/30
5/30
13/30
11/30
25/30
16/30
30/30
24/30
30/30
0/30
13/30
1/30
7/30
29/30
3/30
5/30
26/30
17/30
8/30
30/30
0/30
0/30
0/30
0/30
17/30
0/30
2/30
22/30
23/30
27/30

57
47
46
57
40
31
29
64
19
102
40
28
66
74
11
35
81
33
65
28
35
41
14
109
37
7
60
40
78
18

3.31
3.30
12.06
6.48
7.22
2.57
0.14
7.80

11.42
2.05
4.62
0.44
1.20

1.40
6.06
4.94
0.98
10.49
6.33

4.61
2.47
0.87

11.63
3.35
0.28
0.60

2.50

7.53/30

5.5

0.61

17.3/30

46

3.89

GA-1
Found

Excess (%)

GA-2
Found

Time (s)

Excess (%)

MA
Found

Time (s)

Excess (%)

Found

Time (s)

0/30
0/30
0/30
0/30
0/30
0/30
18/30
0/30
30/30
0/30
0/30
0/30
0/30
0/30
30/30
0/30
1/30
2/30
1/30
1/30
0/30
30/30
0/30
0/30
3/30
30/30
0/30
0/30
0/30
0/30

1.20
0.20
0.36
0.70
0.49
0.06
0.30
1.21

2.45
0.67
3.31
5.75
0.05
0.19
0.31
2.73
0.03
0.00

0.17
0.28

1.70
0.05

0.79

1.39

0/30
26/30
26/30
20/30
2/30
29/30
20/30
6/30
30/30
10/30
20/30
1/30
0/30
27/30
27/30
18/30
0/30
28/30
29/30
30/30
0/30
27/30
30/30
17/30
28/30
30/30
22/30
30/30
15/30
30/30

7.12
7.88
6.01
10.47
7.59
5.10
7.17
7.72
3.51
10.25
6.59
5.39
9.37
12.42
2.71
7.22
13.56
7.72
8.66
6.30
6.37
6.18
4.82
10.69
6.52
4.15
9.18
6.87
10.84
6.29

0.88
0.36

0.40

1.46

1.24

0.32
0.90
0.09

1.24
0.03
0.00

0.13

0.30
0.01
0.16

6/30
9/30
30/30
30/30
8/30
30/30
30/30
1/30
30/30
17/30
30/30
25/30
20/30
11/30
30/30
30/30
1/30
28/30
24/30
30/30
4/30
30/30
30/30
30/30
30/30
30/30
27/30
29/30
27/30
30/30

12.82
20.18
9.76
15.28
13.15
9.61
9.20
13.24
6.71
21.84
14.14
15.05
18.57
25.55
4.12
13.09
30.19
14.37
16.17
10.57
10.29
10.61
8.32
12.49
12.82
6.37
15.26
9.85
23.69
8.10

0.59
0.36

0.40

1.02

0.74

0.30
0.90
0.02

0.81
0.03

0.11

0.30
0.01
0.11

10/30
14/30
30/30
30/30
8/30
30/30
30/30
8/30
30/30
17/30
30/30
25/30
20/30
11/30
30/30
30/30
9/30
28/30
24/30
30/30
10/30
30/30
30/30
30/30
30/30
30/30
29/30
29/30
29/30
30/30

5.24
7.2
3.16
4.45
5.5
7.61
3.2
9.41
2.19
9.24
8.1
5.01
8.13
8.52
1.42
7.32
5.09
8.17
8.03
9.43
4.12
5.61
4.32
6.49
5.2
6.03
5.26
5.51
7.52
4.1

5.23/30

0.81

19.27/30

7.49

0.25

22.9/30

13.71

0.19

24/30

6.02

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

Table 5
MA results for MEB problem instances with 50 nodes.

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

The result of MA for the MEB problem instance with 20 nodes

[15] is compared with ELS, NP, ILO, GA-1 and GA-2 [1] is given in
Table 4 to show its efciency for the benchmark problem. Each
instance is executed for 30 times and the average result is shown
in each row of Table 5. There is no excess value for MA, GA-1 and
GA-2, since the optimal solution is identied for all the 30
instances. GA-1 performs 1-shrink operation, GA-2 and MA performs both 1-shrink and 2-shrink operations. Hence, GA-2 has
increase in computation time but MA reduces the computation
time compared to GA-1 and GA-2.

Fig. 3. Power matrix for the 10-node network.

Fig. 4. Percentage improvement of mean broadcast power.

MEB problem instance with 50 nodes is compared with GA-1

and GA-2 is shown in Table 5. The optimal solution is identied for
24 problem instances. When analyzing the excess percentage it is
found that the solution identied for the remaining instances is
also much nearer to the optimal solution. The computation time is
also reduced compared to GA-1 and GA-2.
From the results of benchmark problem instances shown in
Tables 35, it is proved that MA performs better compared to other
algorithms in terms of average excess percentage over optimal/best
found solution and the number of times the optimal solution found
out of 30 trials. The average excess percentage by MA is much
reduced compared to the other algorithms. In terms of average
execution time, MA performs well compared to GA-1 and GA-2.
Analyzing on the whole of three parameters, MA performs better in
terms of solution found and in execution time. This is because in GA1 and GA-2 the algorithm spends time in nding the initial population, calculating its tness and then taking the best among them to
do crossover and mutation. After these steps, it performs the decoder
operator that creates another set of population with the root node as
the source node, which is processed by the local search operator. The
populations generated before decoder operator is not taken for local
search, which increases the execution time unreasonably.
The percentage improvement of mean broadcast power of
simulated annealing with sweep (SA-sw) [10], genetic algorithm
with sweep (GA-sw) [14], GA-1, GA-2 and MA for 25, 50, 75 and

Fig. 6. Optimal solution analysis of MA with other algorithms for 50-node

instances.

Fig. 5. Comparison of computation time by MA with other algorithms.

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i

D. Arivudainambi, D. Rekha / Swarm and Evolutionary Computation ()

knowledge about the problem. Simulation results show that MA

reduces the execution time to a large amount when compared to
existing algorithms. The results validate the effectiveness and efciency of MA to this problem.

Acknowledgments
We gratefully acknowledge Department of Science and Technology, INDIA for providing nancial support to carry out this
research work under PURSE scheme. We would like to thank
sincerely the Editor and the Reviewers for their constructive
comments and suggestions that helped us to improve the manuscript signicantly.
References
Fig. 7. Excess percentage comparison of MA with other algorithms for 50-node
network.

Fig. 8. Transmission power convergence graph.

100-node network is shown in Fig. 4. The data compared with MA

in gures and tables are reported in [1], [10] and [14]. The results
conrm the efciency of MA over recently proposed algorithms.
The average computation time taken by MA for 20-node and
50-node network of 30 independent trials is shown in Fig. 5. The
time taken by MA compared to GA-2 and GA-1 is largely reduced.
The optimal solution identied by MA for 50-node benchmark
problem instance is compared with other algorithm in Fig. 6. The
dot in the gure indicates the average number of times solution
found by the algorithm. The excess percentage calculated by MA is
compared with other algorithms in Fig. 7. The transmission power
convergence graph for 10-node, 20-node and 50-node network is
shown in Fig. 8. From the graph, it is identied that MA identies the
optimal solution in reduced generations even for large network.
The simulation results conrm the advantages of MA in terms
of optimal solution and running time. MA achieves in shorter
running time compared with other algorithms. The outcome
validates the effectiveness and efciency of MA for the minimum
energy broadcast problem.
5. Conclusion
In this paper, we discussed knowledge added MA for minimum energy broadcast problem. MA actively aims on improving
solution and is explicitly concerned in exploiting all available

[1] Alok Singh, Wilson Naik Bhukya, A hybrid genetic algorithm for the minimum
energy broadcast problem in wireless ad hoc networks, Applied Soft Computing 11 (2011) 667674.
[2] Arindam K. Das, Robert J. Marks, Mohamed El-Sharkawi, Payman Arabshahi,
Andrew Gray, r-shrink: a heuristic for improving minimum power broadcast
trees in wireless networks, in: Proceedings of IEEE GLOBECOM (2003)
523527.
[3] R. Banos, C. Gil, J. Reca, F.G. Montoya, A Memetic, Algorithm applied to the design
of water distribution networks, Applied Soft Computing 10 (2010) 261266.
[4] Di Yuan, Joanna Bauer, Dag Haugland, Minimum-energy broadcast and multicast in wireless networks: an integer programming approach and improved
heuristic algorithms, Ad Hoc Networks 6 (2008) 696717.
[5] Intae Kang, Radha Poovendran, Iterated local optimization for minimum
energy broadcast, in: Proceedings of third International Symposium on
Modeling and Optimization in Mobile, Ad-Hoc and Wireless Networks 2005,
IEEE Computer Society, 2005, pp. 332341.
[6] Jeffrey E. Wieselthier, Gam D. Nguyen, Anthony Ephremides, On the construction of energy efcient broadcast and multicast trees in wireless networks, in:
Proceedings of the IEEE INFOCOM 2000, IEEE Press, 2000, pp. 585594.
[7] Jeffrey E. Wieselthier, Gam D. Nguyen, Anthony Ephremides, Energy-efcient
broadcast and multicast trees in wireless networks, Mobile Networks and
Applications 7 (2002) 481492.
[8] N. Krasnogor, S. Gustafson, Toward truly memetic memetic algorithms:
discussion and proof of concepts, in: D. Corne, G. Fogel, W. Hart, J. Knowles, N.
Krasnogor, R. Roy, J.E. Smith, A. Tiwari (Eds.), Advances in Nature-Inspired
Computation: The PPSN VII Workshops, PEDAL (Parallel, Emergent and
Distributed Architectures Lab), 2002.
[9] P. Moscato, Memetic Algorithms: A Short Introduction New Ideas in Optimization, McGraw-Hill, UK Maidenhead, UK, England 219234.
[10] Roberto Montemanni, Luca Maria Gambardella, Arindam K. Das, The minimum
power broadcast in wireless networks: a simulated annealing approach, in:
Proceedings of Wireless Communications and Networking Conference, IEEE,
2005, pp. 20572062.
[11] Sameh Al-Shihabi, Peter Merz, Steffen Wolf, Nested partitioning for the
minimum energy broadcast problem, in: Proceedings of Learning and Intelligent Optimization Conference, Springer, LNCS, 2008, pp. 111.
[12] Song Guo, Oliver W.W. Yang, Energy-aware multicasting in wireless ad hoc networks: a survey and discussion, Computer Communications 30 (2007) 21292148.
[13] Steffen Wolf, Peter Merz, Evolutionary local search for the minimum energy
broadcast problem, in: Proceedings of Eighth European Conference on Evolutionary Computation in Combinatorial Optimization, Springer, LNCS, 2008,
pp. 6172.
[14] Xiang Wu, Xinheng Wang, Rui Liu, Solving minimum power broadcast problem in
wireless ad-hoc networks using genetic algorithm, in: Proceedings of Communication Networks and Services Research Conference, 2008, pp. 203207.
[15] http://dag.informatik.uni-kl.de/research/meb/.
[16] Hugo Hernandez, Christian Blum, Minimum energy broadcasting in wireless
sensor networks: An ant colony optimization approach for a realistic antenna
model, Applied Soft Computing 11 (2011) 56845694.
[17] Hugo Hernandez, Christian Blum, Distributed ant colony optimization for minimum energy broadcasting in sensor networks with realistic antennas, Computer
and Mathematics with Applications, 64 (2012) 36833700.
[18] Enan A. Khalil, Baraa A. Attea, Energy-aware evolutionary routing protocol for
dynamic clustering of wireless sensor networks, Swarm and Evolutionary
Computation 1 (2011) 195203.
[19] S. Sengupta, S. Das, Md. Nasir, A.V. Vasilakos, W. Pedrycz, Energy-efcient
differentiated coverage of dynamic objects using an improved evolutionary
multi-objective optimization algorithm with fuzzy-dominance, IEEE Congress
on Evolutionary Computation (2012) 18.
[20] S. Sengupta, S. Das Md. Nasir, B.K. Panigrahi, Multi-objective node deployment
in WSNs: In search of an optimal trade-off among coverage, lifetime, energy
consumption, and connectivity, Engineering Applications of Articial Intelligence, 26 (2013) 405416.

Please cite this article as: D. Arivudainambi, D. Rekha, Memetic algorithm for minimum energy broadcast problem in wireless ad hoc
networks, Swarm and Evolutionary Computation (2013), http://dx.doi.org/10.1016/j.swevo.2013.04.001i